Hi Darren and SPMers,
I posted this message several days ago but nobody replied it. Could
anybody kindly have a look at this question?
>
>Thank you for your replies. The comments are extremely helpful!
>
>But the first question is that: Can I use the Bayesian averaging routine
>in DCM (spm_dcm_average) to average the models across the sessions and
>then the set of DCM parameters derived from the averaged model could
>represent the the individual subject's model property and enter into
>second level statistics?
>
>I am aware that the averaged model(across subjects)can not be used in the
>model comparison, as suggested by Dr.Stephan, but I am not sure if it
>works in such a way? Also, if this method (using the coupling
>parameters derived from the averaged model) is feasible, what do you
think of comparing this method with the one computing the arithmetic mean
of the parameter across session-specific models, or comparing with the
method concatenating the data? which one is more appropriate to represent
the model property in the subject level and more powerful when entering
into second level statistics
>
>Best
>
>
>Laurence
>
On Fri, 28 Dec 2007 17:39:17 +0000, <Laurence> <Wang>
<[log in to unmask]> wrote:
>Hi Darren,
>
>Thank you for your replies. The comments are extremely helpful!
>
>But the first question is that: Can I use the Bayesian averaging routine
>in DCM (spm_dcm_average) to average the models across the sessions and
>then the set of DCM parameters derived from the averaged model could
>represent the the individual subject's model property and enter into
>second level statistics?
>
>I am aware that the averaged model(across subjects)can not be used in the
>model comparison, as suggested by Dr.Stephan, but I am not sure if it
>works in such a way? Also, if this method (using the coupling
>parameters from the averaged model) is feasible, what do you think of
>comparing this method with the one computing the arithmetic mean of the
>parameter across session-specific models, or comparing with the method
>concatenating the data? which one is more appropriate to represent the
>model property in the subject level and more powerful when entering into
>second level statistics
>
>Best
>
>
>Laurence
>
>
>On Fri, 28 Dec 2007 09:12:00 -0600, d gitelman <d-
>[log in to unmask]> wrote:
>
>>Hi Laurence:
>>
>>
>>> -----Original Message-----
>>> From: SPM (Statistical Parametric Mapping)
>>> [mailto:[log in to unmask]] On Behalf Of <Laurence> <Wang>
>>> Sent: Thursday, December 27, 2007 10:56 AM
>>> To: [log in to unmask]
>>> Subject: Re: [SPM] DCM with multiple sessions per subject
>>>
>>> Hi Dr. Stephan and SPMer,
>>>
>>> My further question on this issue is that how to characterize
>>> the individual's coupling parameters, if I have a cohort of
>>> subjects and each subject has multiple seesions and now I try
>>> to do the correlation analysis between the coupling
>>> parameters derived from the DCM and behavioral data.
>>>
>>> Actually, I have analysed the data each session separately,
>>> however I am not sure what is the appropriate way to produce
>>> the model parameters representing the individual subject's
>>> property. Could I just average the model (identical model
>>> structure) across the sessions in each subject and consider
>>> the parameters from the averaged model could reflect the
>>> individual's characteristics? or just calculate the
>>> arithmetic mean of the parameters across session-specific
>>> models in each subject?
>>
>>I am not sure what you are averaging at the model level. If it is the
>>parameters then you've answered your second question. You can average the
>>parameters across sessions for identical models.
>>
>>>
>>> I read through the posting and it seems like that you experts
>>> thought it is better to concatenate the data from different
>>> sessions so that it can generate a subject-specific model
>>> under such a condition. Am I correct?
>>
>>I think the reasons that concatenation has been recommended are
>> - you get one set of DCM parameters per model rather than separate sets
>for
>>each session
>> - it is more likely that you will have adequate numbers of trials/blocks
>>across an entire run vs. in individual sessions. For example, you might
>have
>>20 trials of some stimulus across all your sessions, but individual
>sessions
>>might only have 2-5 trials of that type. Having only a few trials in a
run
>>might not produce good estimates of the parameters, but I suppose this
>would
>>also be true for the standard statistics.
>>
>>In any case, assuming an adequate number of trials per session there
>should
>>be no particular barrier to running the DCM's on individual sessions and
>>then averaging the parameters.
>>
>>Hope this helps,
>>Darren
>>
>>
>>>
>>> I would really appreciate your help!
>>>
>>> Laurence
>>>
>>=========================================================================
>========================================================================
|